Past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints

Transcription

1 Climate Dynamics (26) 26: DOI 1.17/s ORIGINAL ARTICLE V. Masson-Delmotte Æ M. Kageyama Æ P. Braconnot S. Charbit Æ G. Krinner Æ C. Ritz Æ E. Guilyardi J. Jouzel Æ A. Abe-Ouchi Æ M. Crucifix R. M. Gladstone Æ C. D. Hewitt Æ A. Kitoh A. N. LeGrande Æ O. Marti Æ U. Merkel Æ T. Motoi R. Ohgaito Æ B. Otto-Bliesner Æ W. R. Peltier I. Ross Æ P. J. Valdes Æ G. Vettoretti Æ S. L. Weber F. Wolk Æ Y. YU Past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints Received: 31 May 25 / Accepted: 22 September 25 / Published online: 2 December 25 Ó Springer-Verlag 25 Abstract Climate model simulations available from the PMIP1, PMIP2 and CMIP (IPCC-AR4) intercomparison projects for past and future climate change simulations are examined in terms of polar temperature changes in comparison to global temperature changes and with respect to pre-industrial reference simulations. For the mid-holocene (MH, 6, years ago), the models are forced by changes in the Earth s orbital V. Masson-Delmotte (&) Æ M. Kageyama Æ P. Braconnot S. Charbit Æ E. Guilyardi Æ J. Jouzel Æ O. Marti Laboratoire des Sciences du Climat et de l Environnement, (LSCE/IPSL, UMR CEA-CNRS 1572) L Orme des Merisiers, Baˆ timent 71, CEA Saclay, Gif-sur-Yvette Cedex, France valerie.masson@cea.fr G. Krinner Æ C. Ritz Laboratoire de Glaciologie et de Ge ophysique de l Environnement, (UMR 5183 CNRS-UJF), Domaine Universitaire, St Martin d Hères, France A. Abe-Ouchi Center for Climate System Research, The University of Tokyo, Kashiwa, Japan M. Crucifix Æ C. D. Hewitt Hadley Centre for Climate Prediction and Research, Met Office, FitzRoy Road, Exeter, EX1 3 PB Devon, UK R. M. Gladstone Æ I. Ross Æ P. J. Valdes School of Geographical Sciences, University of Bristol, University Road, Bristol, BS8 1SS UK A. Kitoh Æ T. Motoi Climate Research Department, Meteorological Research Institute, 1-1 Nagamine, Tsukuba, Ibaraki, Japan A. N. LeGrande NASA Goddard Institute for Space Studies and Center for Climate Systems Research, Columbia University, New York, NY, USA parameters. The MH PMIP1 atmosphere-only simulations conducted with sea surface temperatures fixed to modern conditions show no MH consistent response for the poles, whereas the new PMIP2 coupled atmosphere ocean climate models systematically simulate a significant MH warming both for Greenland (but smaller than ice-core based estimates) and Antarctica (consistent with the range of ice-core based range). In both PMIP1 and U. Merkel IFM-GEOMAR, Duesternbrooker Weg 2, 2415 Kiel, Germany R. Ohgaito Æ A. Abe-Ouchi Frontier Research Center for Global Change (FRCGC), JAMSTEC, Yokohama City, Japan B. Otto-Bilesner Climate Change Research, National Center for Atmospheric Research, 185 Table Mesa Drive, P.O. Box 3, Boulder, CO, 837 USA W. R. Peltier Æ G. Vettoretti Department of Physics, University of Toronto, 6 St. George Street, Toronto, ON, M5S 1A7 Canada S. L. Weber Climate Variability Research, Royal Netherlands Meteorological Institute (KNMI), P.O. Box 21, 373 AE De Bilt, The Netherlands F. Wolk Institut d Astronomie et de Ge ophysique G. Lemaıˆ tre, Universite catholique de Louvain, Chemin du cyclotron, 2, 1348 Louvain-la-Neuve, Belgium Y. Yu LASG, Institute of Atmospheric Physics, Chinese Academy of Sciences, P.O. Box 984, Beijing, 129 People s Republic of China

2 514 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints PMIP2, the MH annual mean changes in global temperature are negligible, consistent with the MH orbital forcing. The simulated last glacial maximum (LGM, 21, years ago) to pre-industrial change in global mean temperature ranges between 3 and 7 C in PMIP1 and PMIP2 model runs, similar to the range of temperature change expected from a quadrupling of atmospheric CO 2 concentrations in the CMIP simulations. Both LGM and future climate simulations are associated with a polar amplification of climate change. The range of glacial polar amplification in Greenland is strongly dependent on the ice sheet elevation changes prescribed to the climate models. All PMIP2 simulations systematically underestimate the reconstructed glacial interglacial Greenland temperature change, while some of the simulations do capture the reconstructed glacial interglacial Antarctic temperature change. Uncertainties in the prescribed central ice cap elevation cannot account for the temperature change underestimation by climate models. The variety of climate model sensitivities enables the exploration of the relative changes in polar temperature with respect to changes in global temperatures. Simulated changes of polar temperatures are strongly related to changes in simulated global temperatures for both future and LGM climates, confirming that ice-core-based reconstructions provide quantitative insights on global climate changes. 1 Introduction The Arctic (Corell et al. 24) and the Antarctic Peninsula (Jacka and Budd 1998; Vaughan et al. 23) are experiencing some of the most rapid temperature rises on Earth. Since the 196s, temperatures have risen by more than 2 C in some areas, several times more rapidly than the average global temperature (IPCC 21; Moritz et al. 22). Due to possible changes in snow and ice cover and associated albedo feedbacks, polar regions are likely to act as amplifiers of climate change, however modulated by local processes generating multi-decadal variability (Bengtsson et al. 24; Polyakov et al. 22). Future climate change simulations in the Arctic indeed show a polar amplification with times the global warming (Holland and Bitz 23). These contrasted results apparently result from three main processes in the climate models: different changes in cloud cover (and cloud radiative feedbacks); different changes in heat advection (Alexeev et al. 25); and, most important, different response in sea ice. The level of future polar amplification was indeed shown to be dependent on the modern sea ice thickness and snow cover simulations (Dixon et al. 23; Holland and Bitz 23). Early studies have revealed that models with cold modern biases, when forced with increased greenhouse gas concentrations, had a larger sensitivity due to a larger ice feedback (Dixon et al. 23; Rind et al. 1997). Reducing the uncertainties of future climate change predictions is extremely important for high latitude areas because (1) local eco-systems will be extremely vulnerable to large warmings because of competition with living organisms from lower latitudes more adapted to warm conditions, and impossibility of migration to colder places (Corell and coauthors 24); (2) ice sheets are susceptible to respond to a fast warming (Rignot and Thomas 22), either by increased accumulation (Huybrechts et al. 24) or coastal thinning (Krabill et al. 24) and modify the freshwater supply to the high latitude oceans; (3) the warming of the Arctic may increase the release of carbon from soils and act as a positive feedback on climate change (Knorr et al. 25). In order to assess the capacity of state-of-the-art climate models to simulate realistically the high latitude climate changes, past climate simulations are compared here with ice-core estimates for last glacial maximum (21, years ago, hereafter LGM) and mid-holocene (6, years ago, hereafter MH) temperature changes. These simulations enable the test of the model responses to a variety of forcings. MH boundary conditions include pre-industrial greenhouse gas levels together with modified orbital parameters (Berger 1978) which induce small changes in low to high latitude annual mean insolation (effect of obliquity change), but with a seasonal cycle increased (respectively reduced by locally 5%) in the northern (respectively southern) hemisphere (in response to precession changes). In this respect, polar MH simulations provide a test of climate model response mainly to a modified seasonal cycle forcing (Renssen et al. 25a, b). LGM boundary conditions include reduced atmospheric greenhouse gas levels (CO 2, CH 4 and N 2 O), modified surface boundary conditions including the presence of massive ice sheets over northern North America and Fennoscandia (Peltier 1994, 24) and small changes in orbital parameters (Table 1). Climate models offer unique frames to understand the spatial representativity of polar ice core sites. In particular, glacial interglacial Antarctic temperature changes have been used to estimate paleoclimatic constraints on climate sensitivity, assuming that they are representative of global temperature changes (Genthon et al. 1987; Lorius et al. 199; Manabe and Broccoli 1985). In this paper, we review the global and polar temperature response of new past and future climate simulations conducted with state-of-the-art coupled climate models with standardized forcings. We have used the database of Coupled Models Intercomparison Project (CMIP, database as of ) and second phase of the Paleoclimate Modelling Intercomparison Project (PMIP2, database of ) (Harrison et al. 22). These recent model intercomparison results are currently used as part of the fourth IPCC assessment. They are also compared to earlier simulations conducted with atmosphere alone models (Paleoclimate Modelling Intercomparison Project, PMIP, Joussaume and Taylor 1995).

3 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints 515 Table 1 Boundary conditions used for the past and future climate simulations Simulations Atmospheric composition (greenhouse gases) Land surface Ocean surface Orbital parameters Years analysed 2 CO2 1 ppm/year increase until CO2 doubling (7 years) then stabilization Control run Pre-industrial 28 ppm CO2 4 CO2 1 ppm/year increase until CO2 quadrupling (14 years) then stabilization Modern Calculated Modern Years First 5 years of the simulations Modern Calculated Modern Years Control run Pre-industrial First 5 years of 28 ppm CO2 the simulations PMIP1 MH fixed SST fix Pre-industrial Modern Fixed to modern 6, BP All years in database Control run Modern Modern (equilibrium simulations) PMIP2 MHCoupled models cpl Pre-industrial Modern or interactively calculated (OAV models) Control run Modern PMIP1 LGM fixed SST fix Glacial 2 ppmv Vegetation : modern, Ice sheets, coastline : (Peltier 1994) Control run Modern Modern Modern Modern PMIP1 LGM Computed SST slab Glacial 2 ppmv Vegetation: modern, Ice sheets, coastline : (Peltier 1994) Control run Modern Modern Modern Post-PMIP1 LGM Coupled models cpl Control run Modern except HadCM3 (Pre-industrial) PMIP2 LGM Coupled models cpl Glacial 185 ppmv Vegetation: control, Ice sheets, (Peltier 1994, 24) Coastline : either unchanged or 15 m Glacial 185 ppmv Vegetation: control, Coastline : 15 m, Ice sheets, (Peltier 24) Calculated 6, BP All years in database (equilibrium simulations) Fixed to (CLIMAP 1981) 21, BP All years in database (equilibrium simulations) Slab ocean 21, BP All years in database (equilibrium simulations) Calculated 21, BP All years in database (equilibrium simulations) Modern Modern Control run Pre-industrial Modern Modern Calculated 21, BP All years in database (equilibrium simulations) For the description of the different models please refer to the CMIP and PMIP database description (

4 516 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints The key questions we address are: what is the performance of the models in simulating past climate changes at the poles? Is there a polar amplification under different forcings? How do relationships between global and polar temperature changes vary for radically different climates such as the LGM and increased CO 2 experiments? 2 LGM and MH climate change: polar ice core data Deep polar ice cores offer unique archives of past climate and environmental changes extending over several climatic cycles in East Antarctica, at Vostok (Petit et al. 1999), Dome F (Watanabe et al. 23) and Dome C (EPICA-community-members 24) and almost a full climatic cycle in Greenland at Summit (Cuffey and Clow 1997; Johnsen et al. 1997) and even further at North- GRIP (NorthGRIP-community-members 24). The progressive distillation of air masses during the moisture advection from mid to high latitudes is classically used to estimate past temperature changes from polar ice core water isotopic composition in either d 18 O or dd (Dansgaard 1964; Lorius et al. 1969). This isotopic thermometer can however be biased by past changes in snowfall seasonal deposition (Krinner et al. 1997; Werner et al. 2); moisture origin (e.g. Boyle 1997; Cuffey and Vimeux 21a, b) or other factors (inversion strength, short term covariance between precipitation and temperature...). The full isotopic composition of snow and ice itself brings constraints on changes in moisture origin, owing to the deuterium excess parameter (d=dd 8d 18 O) (Masson-Delmotte et al. 25a, b; Stenni et al. 21). Climate models are useful tools to help estimate past changes in deposition conditions (seasonal cycle of precipitation, condensation temperature versus surface temperature) which may alter the water stable isotope to temperature relationships. In Antarctica, water stable isotopes only have been used to quantify glacial-interglacial temperature changes. The consistency of ice core dating with an ice flow model fed by the temperature and accumulation derived from water stable isotope fluctuations warrants the realism of these reconstructions, within 2 3% (Jouzel et al. 23; Parrenin et al. 21; Stenni et al. 21; Watanabe et al. 23). LGM to present surface air temperature change is estimated to be 9±2 C; this error bar is estimated by modelling of precipitation isotopic composition changes with an air mass distillation model, taking into account changes in moisture origin derived from measurements of deuterium excess in the ice (Jouzel et al. 23; Stenni et al. 21; Watanabe et al. 23). This temperature change may partly arise from an ice sheet elevation difference between modern and glacial conditions, with variable ranges of estimates (Ritz et al. 21; Peltier 24). Such elevation effects may also partly influence local Antarctic MH temperature change; most inland ice core isotopic records suggest that MH temperatures were either similar or warmer by at most.8 C (Masson et al. 2). A few coastal records such as Taylor Dome or Law Dome suggest that MH temperatures could have been warmer by more than 1 C (Steig et al. 1998). Although associated with 2 3% uncertainty in Antarctica, temperatures derived from water stable isotope measurements in Greenland have been showed to underestimate glacial temperature changes in Greenland. Alternative paleothermometry methods have been developed and are particularly suited for Greenland where the accumulation rate is typically two to ten times larger than for central Antarctica. Precise measurements of the vertical temperature profiles in the boreholes when the drilling operations are completed provide direct reconstructions of past temperatures owing to the heat diffusion in ice sheets. However, the inversion of the heat diffusion equations requires a priori knowledge of the shape of the temperature variations so there are still uncertainties on MH and LGM changes, typically within 2 C (Cuffey et al. 1995; Johnsen et al. 1995; Cuffey and Clow 1997; Dahl-Jensen et al. 1998). These borehole temperature reconstructions show that MH temperatures were typically C warmer than now and LGM temperatures C colder than now. By contrast, temperature reconstructions based on water stable isotopes suggest a smaller MH warming (.5.9 C) (Masson-Delmotte et al. 25a, b) and only a 1 C amplitude for the glacial interglacial change. This can only be reconciled with the LGM borehole temperature profile when taking into account dramatic changes in the precipitation seasonal cycle: the lack of winter snowfall during glacial times, in response to changes in sea ice and land ice cover, could be responsible for a seasonal bias as suggested by simulations conducted with atmospheric general circulation models (Krinner et al. 1997; Krinner and Werner 23; Werner et al. 2). Although obtained at individual locations, the most reliable ice core based temperature estimates (borehole data for Greenland, stable isotopes for East Antarctica) are representative not only of their local drilling site but of central ice caps, as evidenced by the convergent records from a variety of ice cores (Fig. 1). In the following sections, we compare model results averaged over inland ice caps with these ice core based estimates. 3 Model experimental design For future climate change we consider the two scenarios where atmospheric CO 2 levels were respectively stabilised at a doubling of pre-industrial levels (hereafter 2 CO 2 ) and a quadrupling of pre-industrial levels (hereafter 4 CO 2 ) after a 1% per year increase from preindustrial levels. We only consider recent model simulations performed for the IPCC fourth assessment available from the CMIP database. In these two cases, we have calculated the difference between a 5-year period of the stabilisation period (centred on the time period of 75 years after the end of atmospheric CO 2 increase) and a 5-year period of the control run

5 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints 517 Fig. 1 Maps showing the location of ice core records used to infer estimates of LGM temperature changes in a Greenland and b Antarctica. From south to north, in Greenland, open circles locate the sites of Dye 3, Summit (GRIP and GISP2) and NorthGRIP ice cores. In Antarctica, from 9 W in the trigonometric directions, open circles correspond to Byrd, Siple Dome, Taylor Dome, Dome C, Vostok, Komsomolskaia, Dome B and Dome F ice cores (Table 1). For a consistent comparison with the paleoclimate results, we need to compute the response of the climate models closest to the equilibrium response, while keeping a maximum of models for our analysis. A set of ten model results was available for the 2 CO 2 experiment (nine for the 4 CO 2 experiment), including several simulations conducted either with different initial conditions or different model versions from the same groups (Table 2). For past climates, we have used simulations performed with the same coupled model ( cpl ) versions used in CMIP, as part of PMIP2, with a total of seven models for MH and five for LGM. Earlier simulations conducted with three coupled models but with forcings different from the standard protocol of PMIP2 have also been included (post-pmip1) (Table 1). In addition, we have also considered atmospheric alone simulations ( fix ) of the first phase of PMIP (16 models for MH and

6 518 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints Table 2 List of simulations and model versions used for this intercomparison Model name in the database PMIP1 MH Fixed SST PMIP2 MH Coupled models PMIP1 LGM Fixed SST PMIP1 LGM Calculated SST Post-PMIP1 LGM Coupled models PMIP2 LGM Coupled models CMIP 2 CO2 Coupled models CMIP 4 CO2 Coupled models bmrc X ccc2. X X X ccm1 X ccm3 X ccsr1 X X csiro X climber-2 X cnrm-2 X echam3 X X gfdl X X gen1 X giss-iip X lmcelmd4 X X X lmcelmd5 X X mri2 X X X ugamp X X X uiuc11 X ukmo X X yonu X MIROC3.2 (medres) X X X X NCAR_CCSM3 X X X X FOAM X IPSL-CM4-V1-MR X X X X UBRIS-HADCM3M2 X OA and OAV GISSmodelE-R X X UKMO-HADCM3M2 X ECBILTCLIO X MRI-CGCM1 X HADCM3 X UToronto-CCSM1.4 X CNRM-CM3 X GFDL-CM2. X X GFDL-CM2.1 X X GISS-EH X GISS-ER X X FGOALS-g1. X X X (3 simulations) INM-CM3. X X ECHAM5/MPI-OM X X MRI-CGCM2.3.2 X X (2 simulations) Number of simulations For the description of the different models please refer to the CMIP and PMIP databases descriptions (

7 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints for LGM) as well as nine atmospheric models coupled with slab ocean models ( cal ) for LGM (Tables 1, 2). The fix and cal PMIP runs are equilibrium simulations and all years of simulated climate available in the database were averaged (Table 1); for PMIP2, modelling groups provided 3 to 1 year long outputs, obtained after major adjustments to the forcings (but without constraints on deep ocean equilibrium). These 1 year long outputs where averaged for our analyses. The most recent simulations (CMIP and PMIP2) use coupled ocean atmosphere models and their reference simulations are long pre-industrial control runs. In the case of PMIP1 atmosphere and atmosphere with slab ocean models, the reference simulation is the modern climate. In between these two standardized procedures, post-pmip1 simulations used either modern or preindustrial reference simulations; these differences are not critical for our analyses. We have extracted from the CMIP, PMIP1 and PMIP2 databases the available monthly model outputs (temperature, orography and for the paleoclimate simulations also precipitation). In all the figures and the rest of this paper, all the temperature anomalies will be discussed as control minus LGM, control minus MH, and 2 CO 2 or 4 CO 2 climate minus control climate, for consistency of time direction. Global and regional results were calculated by weighting each model grid point by the actual area of these grid points. For each available simulation, we have calculated the global mean temperature change. For each model, Central Greenland was defined as the model grid points located between 65 and 8 N, 6 and 25 W, with an elevation above 1,3 m. The same approach was used for Central Antarctica with grid points between 9 and 7 S, and 18 E, and an elevation above 2,5 m. 4 Results and discussion 4.1 Model-data comparisons: preliminary remarks Last glacial maximum ice cap elevations The representation of the orography in atmospheric models strongly depends on their resolution and their discretisation method (spectral or grid-point). The average control run central Greenland orography varies between 1,315 and 2,376 m. This dispersion could explain the large range of model results for the control simulation. We therefore do not compare absolute values for each period but anomalies with respect to the control runs. The ICE-4G LGM ice cap prescribed in PMIP1 and post-pmip1 (Peltier 1994) is characterized by an average 8 m elevation increase compared to the present altitude (between 64 and 924 m). In contrast, the ICE-5G LGM ice cap reconstruction prescribed for the PMIP2 simulations (Peltier 24) have a much reduced central Greenland elevation anomaly ( m). However, both ice-core air content information (Raynaud et al. 1997) and ice sheet modelling (Tarasov and Peltier 23) suggest that changes in central Greenland elevation were limited (and even in the opposite direction). We therefore face the problem that different LGM simulations were conducted with different glacial Greenland elevation changes, significantly larger than those derived from ice core information and ice sheet modelling. When considering the typical polar vertical temperature lapse rate (circa 1 C per 1, m) (Krinner and Genthon, 1999), such discrepancies in elevation changes (up to 8 m) may account for up to 8 C temperature differences (without taking into account atmospheric dynamics related to glacial ice cap topographies). In central Antarctica, the modern elevation represented in the models varies between 2,956 and 3,448 m. For PMIP1, the prescribed ice sheet change (Peltier 1994) results in an average elevation increase of 368 m (from 12 to 43 m depending on the models). For post- PMIP1, all the models analysed here show changes of elevation between 392 and 399 m. For PMIP2 with the ICE-5G ice caps, similar orders of magnitude are also obtained (Peltier 24) (elevation changes ranging between 343 and 4 m above modern levels). Again, this central Antarctic ice cap elevation increase at the LGM is too high when compared with ice core air content changes or with dynamical ice cap modelling forced by ice-core based climate histories (Ritz et al. 21). Because of the hundreds of meters of maximum differences both in modern polar elevations and in LGM to modern changes, we have compared model outputs directly, as well as with a correction for changes in elevation. Observations of modern polar vertical lapse rate and spatial elevation-temperature gradients show lapse rates varying between 6 C per 1, m and 15 C per 1, m depending on the season and the location (Ohmura and Reeh 1991; Krinner and Genthon 1998, 1999). In order to account for large glacial to modern central ice sheet elevation changes, we have used a constant first order lapse rate of 1 C per 1, m. The use of this lapse rate provides a rough estimate of the effect of the prescribed orography changes; we estimate that this elevation correction is associated with a maximum uncertainty of 5%, given the range of observed modern polar lapse rates. For reference, we have applied this methodology to compare the control simulations (Table 1) with the modern temperatures at deep ice core sites. The range of control simulation elevations are displayed together with temperatures uplifted at the elevation of ice core sites (typically 3, m for central Greenland and 3,4 m for East Antarctic Plateau). It can be observed that, in average, climate models capture the correct expected range of elevation-corrected temperatures in central Greenland. By contrast, there is a tendency to observe an underestimation of elevation-corrected temperatures on the East Antarctic Plateau by a few degrees (Table 3); this may be due to different representations of the Antarctic inversion strength.

8 52 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints Table 3 «Central Greenland» and «East Antarctic Plateau» control simulations of elevation and temperature respectively uplifted to 3, and 3,4 m elevation (taking into account a vertical lapse rate of 1 C per 1, m) Control runs for experiments Central Greenland corrected temperature ( uplifted at 3, m elevation, C) Central Greenland elevation (m) East Antarctic Plateau corrected temperature ( uplifted at 3,4 m elevation, C) East Antarctic Plateau simulated / observed elevation (m) PMIP1 6fix 29.6±5.7 1,921± ±8. 3,95±12 PMIP1 21 fix 28.4±8.2 1,942± ± ±7 PMIP1 21 cal 3.5±4.3 1,899± ±7.6 3,115±85 Pre-PMIP2 32.9±5.4 1,835± ±7.2 3,59±3 PMIP2 6 k cpl 34.1±2.9 1,826± ±3. 3,53±72 PMIP2 21 k cpl 33.5±3.4 1,883± ±5.5 3,57±71 IPCC 2 CO2 32.1±2.9 2,7± ±2.7 3,5±41 IPCC 4 CO2 32.±3.2 2,5±15 5.4±2.5 3,65±24 GRIP ,2 NorthGRIP ,917 Vostok ,49 Dome C ,2 Dome B 55. 3,65 Komsomolskaia ,5 For comparison, mean characteristics of various drilling sites (see Fig. 1) are also displayed. Uncertainties in the vertical lapse rate used for elevation corrections should induce uncertainties on corrected temperatures within 1 2 C. For each type of control simulation, the mean value of all model outputs available is displayed together with the inter-model standard deviation Comparing model outputs with ice core isotope paleothermometry Water stable isotopes record past temperature changes through the progressive depletion of atmospheric vapour of heavy molecules along their transport path from moisture sources to the poles. The relationship between isotopic composition and temperature therefore results from the control of cloud temperature on condensation. Changes in moisture source origin, in the gradient between surface and condensation temperatures may induce changes in the temporal relationships between polar precipitation isotopic composition and temperature. Deuterium excess data from both poles suggest that changes in moisture source is not a major bias on glacial interglacial ice core based temperature reconstructions (Cuffey and Vimeux 21a, b; Stenni et al. 23, 21). By contrast, changes in the seasonal cycle of precipitation remain one plausible explanation for the mismatch between Greenland glacial temperature change derived from borehole measurements versus temperature derived from the classical spatial isotope/temperature slope (Masson-Delmotte et al. 25, b). In the present work, we have not assessed changes in the vertical temperature gradient, some of the databases which we use missing this type of data. However, we have attempted to estimate the impact of precipitation seasonality changes on past temperatures recorded in ice cores. For this purpose, we have taken into account the impact of changes in the seasonal cycle of the precipitation by comparing the change in annual mean temperature (arithmetic average of 12 monthly mean temperatures) with the change in precipitation-weighted annual mean temperature, for each LGM simulation (calculated from monthly outputs of temperature and precipitation; Fig. 4a, b), following the methodology of (Krinner et al. 1997; Krinner and Werner 23; Werner et al. 21). When calculated from all the control simulations available, this seasonality effect accounts for a 2.3±1.7 and 3.3±2.9 C difference between the modern precipitation weighted temperatures and the annual mean temperatures, respectively for Greenland and East Antarctica. This is the consequence from the fact that climate models simulate slightly more precipitation today during the warmest months in both polar regions than during the coldest months; this effect is reinforced at the LGM. Indeed, PMIP2 results show systematic shifts towards more precipitation provided during the warm season at LGM. If this change of seasonal cycle is real, this effect may account for an underestimation of LGM annual mean temperature changes as recorded in stable isotopes from ice cores, which remains however limited. The differences between weighted LGM and annual mean LGM temperatures increase to 4.±.8 and 4.6±1.3 C for Greenland and Antarctica, accounting for a relative uncertainty of 8 23% of the full glacial-control range, depending on the hemisphere and on the model. 4.2 Global mean temperature changes In order to assess the range of global temperature changes simulated by climate models under 2 CO 2, 4 CO 2, MH and LGM boundary conditions, we have compared on Fig. 2 the distribution statistics (full range, 25th, 5th, 75th quartiles and the mean of all models) of model results for these various experiments. This presentation is well suited when limited numbers of simulations are available (in our case from 4 to 16) and helps visualizing the range (maximum to minimum values), the average, median and dispersion of model results. In particular, extrema levels significantly different from the

9 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints 521 Fig. 2 Annual mean global temperature changes simulated by a variety of climate models run under similar boundary conditions. The full range (dashed line), mean (cross symbol), 25th (lower bold dash symbol), 5th (grey square symbol), 75th (upper bold dash symbol) percentiles of the various model results are calculated from the distribution of the various model results (see Table 2). The fix, slab and cpl abbreviations refer to different configurations of models used and are described in Sect. 3 and Table 1 Global temperature change ( C) th 5th 25th mean 1 PMIP1 PMIP2 IPCC IPCC PMIP1 PMIP1 pre-pmip2 PMIP2-6 cpl -6 cpl 2xCO2-4xCO fix -21 cal -21 cal -21 cpl 25th to 75th ranges clearly identify the occurrence of outliers among the different model results. For the MH, 6, years ago, climate models do capture reconstructed seasonal and regional changes such as wetter conditions in northern Africa (Joussaume et al. 1999; Braconnot et al. 1999) or warmer conditions in some areas of the mid-latitudes in summer (Cheddadi et al. 1997; Masson et al. 1999; Wohlfahrt et al. 24). This results from a seasonal and latitudinal redistribution of solar radiative forcing due to changes in precession and obliquity of the Earth s orbit (Berger 1978; Joussaume and Braconnot 1997) modulated by the differential heat capacity of the ocean and land masses as well as by the sea ice response at high latitudes (Crucifix et al. 22). In some regions, the seasonal changes do have a signature in the annual mean; this is for instance the case for annual mean precipitation in monsoon regions. Globally, the mean temperature change between control and MH simulations is between.5 to +.4 C with a median value of respectively +.5 and +.8 C for PMIP1 atmospheric models and PMIP2 coupled climate models. In average, climate models simulate slightly warmer control global temperatures compared to the MH simulations (Fig. 2). Note that in this MH case, when global mean temperature changes are small, the notion of global climate sensitivity computed from global and annual averages is not suitable because it does not account for the seasonal aspects both in terms of orbital forcing and climate response. The median range of control minus LGM global temperature change is fairly constant for PMIP1 simulations conducted with CLIMAP ocean surface boundary conditions ( fix ) and atmospheric models alone, PMIP1 simulations with atmospheric models coupled with mixed layer ocean models ( cal ) and PMIP2 simulations with fully coupled ocean-atmosphere climate models ( cpl ). The median of model results is rather stable from PMIP1 to PMIP2 (3.7 C for fixed SST, 4.9 C for calculated SST, 3.9 C for post-pmip1 and 4.1 C for PMIP2 results). With the exception of two outliers (LMCELMD4 for PMIP1 cal and UTORON- TO-CCSM1.4 for post-pmip1), all model results range between 3.1 and 6.4 C in terms of glacial to control global temperature change The range of global temperature change for 4 CO 2 experiments, with a median of 4.3 C, is comparable to the LGM to modern change. In the case of 2 CO 2 experiments, the global temperature range has a median of 2. C. The dispersion range between various model results (indicated by the vertical dashed line in Fig. 2) is significant when compared to the full amplitude of the changes. In the case of future climate change scenarios, climate models have a range of dispersion of the same order of magnitude as the median change: 2.3 C for 2 CO 2 and 2.7 C for 4 CO 2. A similar 2 C dispersion is also observed for PMIP2 LGM to modern changes. The difference among individual models forced with similar boundary conditions is therefore about 1% of the median global temperature change in the case of 2 CO 2, and 5% of the median global temperature change in the case of 4 CO 2 and LGM simulations. This comparison shows that the global temperature response is very similar under 4 CO 2 and LGM forcings, although occurring in response to different spatially-distributed forcings (greenhouse gases in one case, greenhouse gases plus ice sheets in the other case). 4.3 The particular case of MH polar temperature change: model-data comparison For the MH, it is impossible to discuss a polar amplification of climate change, defined as the ratio between

10 522 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints the temperature change in central polar regions to global temperature changes, due to the negligible global mean temperature changes. For central Greenland, simulations conducted with fixed ocean surface conditions within PMIP1 show a large dispersion of temperature change between MH and control conditions (range between.8 and 1.5 C with half of the models simulating warmer and half of the models simulating colder conditions than present). This dispersion is probably related to different responses of the atmospheric stationary waves and the associated heat transport to Greenland; note that the range of temperature changes is quite small compared to the typical temperature interannual variability range (Masson et al. 1998). By contrast, all the PMIP2 coupled model simulations performed with an interactive ocean and sea ice model simulate a temperature decrease from MH to modern conditions varying from.35 to.9 C, consistent with paleoclimatic estimates of warmer MH conditions, but weaker than borehole temperature reconstructions (Dahl-Jensen et al. 1998; Masson-Delmotte et al. 25a, b). The use of fully coupled climate models therefore improves the consistency of the model results over Greenland. Indeed, coupled climate model simulate a shaping of the seasonal cycle by the ocean response associated with a late summer warming, a reduced sea ice cover and an Arctic warming (Vavrus and Harrison 23). Although PMIP2 models do capture the correct sign of the reconstructed temperature change, they underestimate the amplitude of the MH Greenland temperature change in average by a factor of 3. Last, these simulations enable us to explore changes in the precipitation seasonal cycle as a clue to reconcile borehole and water isotope temperature reconstructions. All the PMIP2 MH simulations reveal that precipitationweighted temperature changes should be larger than the arithmetic annual mean of monthly temperatures, resulting in a range of precipitation-weighted temperature changes from 1. to 1.2 C (due to more snowfall during the warmest months). This is not consistent with the different ranges of temperature changes estimated from borehole versus stable isotope thermometry, and further suggests that the change in moisture advection simulated by PMIP2 coupled models may not be correct. It has to be noted that the PMIP2 simulations did not systematically include changes in vegetation. Earlier studies had shown that changes in high northern latitude vegetation enhanced the local warming (Wohlfahrt et al. 24). We have performed the same analysis of MH model results for central Antarctica. PMIP1 simulations show no model consistency and result in changes in Antarctic temperatures varying from.6 to +.7 C. Again, PMIP2 simulations performed with coupled models systematically show a temperature decrease from MH to modern levels, by.1.7 C. As for Greenland, this larger consistency among models is attributed to seasonal ocean and sea ice response in the surrounding areas. For central Antarctica, coupled models are qualitatively and quantitatively in agreement with temperature reconstructions based on water isotopes from polar ice cores, which suggest a range of temperature change from to +.8 C (Masson et al. 2). When precipitation seasonality is considered, the simulated temperature range shifts by.3.6 C, confirming the previous range of temperature change and furthermore the fact that changes of Antarctic precipitation seasonal cycle should not be a major source of uncertainty when reconstructing local past temperature changes from ice core water stable isotopes, even within the current interglacial period (Jouzel et al. 23). 4.4 Polar amplification: LGM versus future climate simulations Here we discuss the ratio of polar versus global mean temperature changes. Figure 3a displays the comparison of LGM and future climate change range of model results for CMIP and PMIP1 and PMIP2 intercomparison exercises. Increasing CO 2 concentrations imposed in 2 CO 2 and 4 CO 2 induce a central Greenland temperature change of respectively 1.37 and 1.4 times above the simulated global temperature increase; in central Antarctica, this amplification factor is respectively 1.16 and This intercomparison therefore confirms previous studies conducted as a function of latitude but only for the northern hemisphere (Holland and Bitz 23). Because the median of all climate models is systematically above 1 (Fig. 3), we can reasonably argue that both past and future climate changes are indeed associated with a polar amplification of climate change. For the LGM, part of this polar amplification may be due to the local elevation increase imposed by ice sheet reconstructions used as boundary conditions (Peltier 1994, 24) (see Sect ). In order to assess the proportion of local temperature changes which could be accounted for by local topography changes, we have added elevation corrected results where LGM simulated temperature changes are corrected from changes in elevation using a slope of 1 C per 1, m elevation (these calculations appear as corr labels in Fig. 3). Glacial boundary conditions induce a very large polar amplification for Greenland (Fig. 3a). PMIP1 boundary conditions result in a factor of 5 between Greenland and global temperature changes, and a factor of 3 when corrected for large elevation changes. Such a large amplification is clearly the result of the imposed CLIMAP surface conditions (including a large sea ice coverage in the North Atlantic plus rather warm tropical temperatures). By contrast, PMIP2 simulations conducted with a different ice sheet topography (Peltier 24) and with fully coupled ocean atmosphere sea ice models suggest a range of amplification of (respectively without and with topography correction) (Fig. 4a). These results suggest

11 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints 523 a th 5th 25th mean ratio of 1 ratio of Greenland to global temperature change slope IPCC IPCC PMIP1 PMIP1corr PMIP1 PMIP1corr pre-pmip2 pre-pmip2corr PMIP2 PMIP2 corr 2xCO2-4xCO fix -21 fix -21 cal -21 cal corr -21 cpl -21 cpl -21 cpl -21 cpl b th 5th 25th mean ratio of 1 2 ratio of Antarctic to global temperature change slope IPCC IPCC PMIP1 PMIP1corr PMIP1 PMIP1corr pre-pmip2 pre-pmip2corr PMIP2 PMIP2 corr 2xCO2-4xCO fix -21 fix -21 cal -21 cal corr -21 cpl -21 cpl -21 cpl -21 cpl Fig. 3 a Central Greenland polar amplification (defined as the ratio between central Greenland and global annual mean temperature changes) simulated by climate models. The full range (dashed line), mean (cross symbol), 25th (lower bold dash symbol), 5th (grey square symbol), 75th (upper bold dash symbol) percentiles of the various model results are calculated from the distribution of the various model results (see Table 2). corr stands for elevationcorrected temperature values (see text). b Same as (a) but for central eastern Antarctica. Note that the vertical scale is half as small as for Greenland that Greenland temperature changes are deeply amplified under glacial conditions, due to (1) imposed local topography changes (up to half of the signal), (2) northern hemisphere albedo, atmospheric and ocean circulation effects. The impact of large sea ice changes in the North Atlantic on Greenland temperature is most obvious from the dispersion of PMIP1 slab ocean simulations. Many of these processes should be active both in the warming from glacial to modern conditions and in the warming from modern to future conditions. Note that future climate change simulations analysed here do not account for the possible polar ice sheet

12 524 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints response in terms of elevation change. Preliminary results obtained with an earth climate model of intermediate complexity fully coupled to Greenland and Antarctica ice-sheet models (S. Charbit et al., personal communication) suggest that such effects should be important for central Greenland, similar to the findings of (Huybrechts et al. 24). In Central Antarctica (Fig. 3b), the range of polar amplification is smaller than for central Greenland both for future and glacial climate changes; this is probably due to the larger albedo effect in the northern hemisphere due to ice caps and snow cover over the continents, and to changes in northern hemisphere storm tracks. PMIP1 boundary conditions result in an amplification factor of 2.1 (without orography correction) to 1.2 (with orography correction). PMIP2 simulations conducted with coupled climate models give rather similar ranges of changes (median model amplification factors of respectively 1.9 and 1.2). The sea ice changes simulated by PMIP1 slab ocean runs, post-pmip1 and PMIP2 models is systematically less extensive than previously prescribed from CLIMAP reconstructions, consistent with new austral sea ice reconstruction efforts (Gersonde et al. 25). When orography effects on central Antarctic temperature changes are roughly accounted for (Fig. 4b), the order of magnitude of polar amplification are very comparable to the range of future Antarctic amplification (around 1.2); some models even show no Antarctic amplification. The coherency of Antarctic temperature change with global temperature change may result from a dominant role of greenhouse gases on southern high latitude radiative budget, only slightly modulated by the internal feedbacks associated with sea ice extent, cloudiness, atmospheric heat and moisture advection (including water vapour feedback, more important at high latitudes). By contrast, we can speculate that Greenland polar amplification should be more influenced by other processes such as dust forcing, changes in vegetation in the northern mid and high latitudes, large icy surfaces and associated feedbacks on atmospheric circulation and heat transport to Greenland. 4.5 Ice-core constraints on global temperature changes When taking into account all available 2 CO 2 and 4 CO 2 simulations, a linear regression relating Greenland to global temperature change leads to a slope of 1.46 and a determination coefficient (R 2 ) of.9 (Fig. 4c). Therefore, in the world of model results, in average, future changes in Greenland temperature are linearly determined by the global temperature range ( climate sensitivity ) but amplified by a factor of 46% ( polar amplification ). When considering glacial to control climate changes, we restrict here our analysis to the coupled model simulations performed under the same boundary conditions and with models involved in the IPCC fourth assessment (PMIP2 simulations). In this case, central Greenland temperature change is also linearly related to global temperature change with a slope of 2.71 (without topography correction) and a determination coefficient of.81 (Fig. 4a). This suggests that paleoclimatic polar temperature reconstruction provide a quantitative constraint on the range of global temperature change. In fact, the range of simulated central Greenland temperature changes lies between 6.2 and 14.5 C, significantly below the borehole temperature estimates of C for 21, years before present. As for the MH climate, coupled model results do capture the glacial cooling but underestimate by a factor from 1.3 to 3.5 the reconstructed temperature change. Part of this inconsistency could be attributed to the uncertainty linked with the effect of increased dust concentrations on climate during glacial times, probably underestimated when taken into account with atmosphere-only models (Werner 22). Other missing factors include the lack of LGM vegetation (Crowley and Baum 1997) and soil (Poutou 23) feedbacks, more important in the northern hemisphere but which should be rather limited for Greenland (Crucifix and Hewitt 25). When extrapolated, this model-data comparison however suggests that the simulations most consistent with the paleoclimatic constraints are associated with large glacial to control global temperature changes (above 4 C) (Fig. 4b). The precipitation seasonality effect remains negligible (within a few C) compared to the model-data discrepancy, and even acts against any reconciliation of the isotope to the borehole thermometry (all model results give a smaller amplitude of changes when precipitation weighted). The consistency of model results could question the interpretation of proxy records; however, borehole temperature profiles in central Greenland are based on a physical mechanism (diffusion of heat into the ice). For central Antarctica, a similar analysis can be performed. When a linear regression is performed on all the range of 2 CO 2 and 4 CO 2 simulations, this leads to a slope of 1.16 and a determination coefficient of.8, suggesting again that future Antarctic temperature change is, in the models, related to the global temperature change, and in average amplified by about 16% above the global mean temperature change (Fig. 4d). When applied to PMIP2 last glacial maximum to control results, the same analysis leads to a slope of 2.9 and a smaller determination coefficient of.7 (Fig. 4b). This result suggests that Antarctic temperature changes are representative of large scale temperature changes (up to the global scale), with an amplification factor of 2. This is a model confirmation that the early hypothesis of (Genthon et al. 1987) and (Lorius et al. 199) remains valid. Antarctic glacial interglacial temperature changes can be considered to constrain global climate sensitivity to greenhouse gases. The range of simulated glacial to control central Antarctic temperature change varies between 2.4 C for

13 Masson-Delmotte et al.:past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints 525 Fig. 4 a Comparison of Last Glacial Maximum to control central Greenland annual mean temperature change simulated by climate models (PMIP2 coupled ocean-atmosphere simulations only) with the range of paleoclimatic reconstructions. Filled black squares show direct model results. Open black squares show model results corrected from LGM to control ice sheet elevation changes ( elevation corrected results). Grey squares show model results corrected from elevation changes and precipitationweighted ( seasonality corrected results). Horizontal long-dashed lines reflect the range of temperature change derived from Greenland borehole thermometry. Shortdashed lines correspond to slopes of 1, 2 and 3 for reference. A linear regression calculated on the results of these four models is also displayed (solid black line and regression result). Values below zero are not displayed (results of ECBILT CLIO with corrections. b Same as (a) but for central Antarctica. Horizontal long-dashed lines reflect the range of temperature change derived from Antarctic ice core water stable isotopes. c Same as (a) but for future climate change simulations. Open black squares represent 4 CO 2 simulation anomalies, and filled black rhomboids 2 CO 2 simulation anomalies. The solid line is a linear regression on all the simulation results. The black dashed lines represent lines with slopes of 1 and 2. d Same as (b) but for future climate change simulations. Open black squares represent 4 CO2 simulation anomalies, and filled black rhomboids 2 CO2 simulation anomalies. The solid line is a linear regression on all the simulation results. The black dashed lines represent lines with slopes of 1 and 2 Greenland temperature change ( C) Antarctic temperature change ( C) a b y = 2.64x R 2 = Global temperature change ( C) HadCM3M2 HadCM3M2 HadCM3M2 MIROC3.2 MIROC3.2 MIROC3.2 CCSM CCSM CCSM IPSLCM4 IPSLCM4 IPSL CM4 ECBILTCLIO ECBILTCLIO slope 1 slope 2 slope 3 range of reconstructions range of reconstructions y = 2.7x R 2 = Global temperature change ( C) HadCM3M2 HadCM3M2 HadCM3M2 MIROC3.2 MIROC3.2 MIROC3.2 CCSM CCSM CCSM IPSLCM4 IPSLCM4 IPSLCM4 ECBILTCLIO slope 1 slope2 slope 3 range of reconstructions range of reconstructions ECBILT-CLIO (an intermediate complexity climate model) to 1.8 C. The impact of elevation changes accounts for C of the glacial to pre-industrial warming. Without orography correction, all of the four general circulation coupled model results fall within the range of ice-core based temperature range (9±2 C).